Today, the lifetime of the SizeCountCheckpointValidator is equal to lifetime of the OplogApplier which is actually equal to the process lifetime. If a node creates an OplogApplier at any point, steps up, and then steps back down, it continues using the state in the pre-existing SizeCountCheckpointValidator. Oplog application stops during the node's primary tenure, so the SizeCountCheckpointValidator state is stale, and subsequent checkpoint validation is fails.

Change the lifetime of the OplogApplier::_checkpointValidator to be one secondary term.